EXGBoost: Gradient Boosting in Elixir(dockyard.com)
dockyard.com
EXGBoost: Gradient Boosting in Elixir
https://dockyard.com/blog/2023/07/18/introducing-exgboost-gradient-boosting-in-elixir
8 comments
Is it just me or has HN been full of Elixir posts recently? Is there a trend I’m missing?
Elixir has always been a well liked language on here and in general. The past year there has been a lot of work on building data and AI/ML libraries in Elixir.
Explorer, Nx, Axon, Scholar, Bumblebee, Broadway, Livebook and now EXGBoost.
You can get pretty close to building a full stack AI application in Elixir now.
From data ingestion and processing all the way to serving the model in a Phoenix web app. All in Elixir.
I think it's great more languages are becoming alternates to Python in data and AI/ML, more options the better.
Explorer, Nx, Axon, Scholar, Bumblebee, Broadway, Livebook and now EXGBoost.
You can get pretty close to building a full stack AI application in Elixir now.
From data ingestion and processing all the way to serving the model in a Phoenix web app. All in Elixir.
I think it's great more languages are becoming alternates to Python in data and AI/ML, more options the better.
There have been a few just recently. Bit of an ML boom in Elixir and ML is popular right now.
As an Elixir-centric blogger I can tell you that Elixir generally seems to do well on Hacker News. I think it has a lot of overlap with interests on here. Erlang is the underpinnings and Erlang is generally interesting from a comp sci standpoint. Elixir's roots in Ruby line it up nicely with SaaS, startups and that whole space.
As an Elixir-centric blogger I can tell you that Elixir generally seems to do well on Hacker News. I think it has a lot of overlap with interests on here. Erlang is the underpinnings and Erlang is generally interesting from a comp sci standpoint. Elixir's roots in Ruby line it up nicely with SaaS, startups and that whole space.
For the uninitiated, what is the advantage of using Elixir for machine learning?
For me it would be, first and foremost, "not having to deal with Python's many idiosyncrasies". Even when I go with best-practices, trying to git clone a Python project and "pip install" something into a virtualenv, it's STILL hit-or-miss whether it conflicts with something else. It's a very janky experience and I don't like it, and this isn't even before touching the language itself, which is well-covered ground already: https://medium.com/nerd-for-tech/python-is-a-bad-programming...
I am only reposting my own tweet here because I just sent an answer to the exact same question to a youtuber who was curious: https://twitter.com/pmarreck/status/1684248660832288771?s=20
It has a few links.
There's also this podcast (which has a transcript link): https://smartlogic.io/podcast/elixir-wizards/s10-e10-sean-mo...
Lastly, I just found https://www.thestackcanary.com/from-python-pytorch-to-elixir... which explains some reasons.
I am only reposting my own tweet here because I just sent an answer to the exact same question to a youtuber who was curious: https://twitter.com/pmarreck/status/1684248660832288771?s=20
It has a few links.
There's also this podcast (which has a transcript link): https://smartlogic.io/podcast/elixir-wizards/s10-e10-sean-mo...
Lastly, I just found https://www.thestackcanary.com/from-python-pytorch-to-elixir... which explains some reasons.
You're in luck, there was a big discussion on that just yesterday.
https://news.ycombinator.com/item?id=36859785
Elixir is good at doing a lot of things at once on - scaling to lots of machines - and not exploding catastrophically while doing so.
Turns out this is really helpful for machine learning where you want to coordinate big data pipelines and do things like batching requests to a GPU resource (because GPUs want to be parallelized).
You can do batched ML inference pretty much out of the box with Nx.Serving https://hexdocs.pm/nx/Nx.Serving.html where you'd have to spin up a separate third party service like https://developer.nvidia.com/triton-inference-server otherwise.
Turns out this is really helpful for machine learning where you want to coordinate big data pipelines and do things like batching requests to a GPU resource (because GPUs want to be parallelized).
You can do batched ML inference pretty much out of the box with Nx.Serving https://hexdocs.pm/nx/Nx.Serving.html where you'd have to spin up a separate third party service like https://developer.nvidia.com/triton-inference-server otherwise.
Awesome! Been looking forward for this release!